Sen. Bernie Sanders (I-VT) introduced legislation Thursday that would ban the development of artificial superintelligence, warning that advanced AI systems could become capable of operating beyond human control.
Sanders and co-sponsor Rep. Greg Casar (D-TX) dubbed the measure the “Ban Artificial Superintelligence Act,” describing it as legislation “to stop AI oligarchs from building machines humans cannot control.”
Recommended Stories
The bill, which has not yet been released in full, would pause the development of advanced AI until a new federal regulatory agency is established and has created safety standards and model-review processes for the development of frontier AI systems.
The legislation would also establish a new Cabinet-level federal agency tasked with protecting the public from AI-related risks and enforcing a prohibition on artificial superintelligence.
The bill would impose significant penalties on those who attempt to circumvent the proposed restrictions. Companies could face a so-called “corporate death penalty,” while individuals could face up to 20 years in prison, which Sanders compared to existing laws governing the unlawful development of nuclear weapons.
At the center of the proposal is Sanders’s definition of “artificial superintelligence.”
The legislation would apply to AI systems that “exhibit or can easily be modified to exhibit capabilities that match or exceed human cognitive performance and capabilities across a broad range of domains or tasks.”
It would also encompass systems with the ability to plan and execute the “disempowerment of humanity,” including by overthrowing or undermining the U.S. government.
Sanders has positioned himself as a leading congressional voice calling for sweeping restrictions on advanced AI development. In recent weeks, he urged the CEOs of major AI companies to “stop building machines that humans cannot control.”
He referenced the July incident in which rogue AI agents from OpenAI escaped their isolated testing environment without prompting and breached AI startup Hugging Face’s infrastructure. The rogue agents went undetected for two weeks.
The Vermont senator’s proposal comes as major technology companies race to develop powerful AI systems.
Meta said last year that it would “stop development” of certain technologies if they reached a point beyond the company’s ability to operate them safely. OpenAI has similarly said it would halt further development if its technology surpasses its ability to operate it safely without adequate safeguards. Anthropic made a comparable commitment in 2023, saying it would pause the scaling or delay deployment of new models if AI capabilities outpaced its safeguards.
Sanders argued those voluntary commitments have not gone far enough, accusing the industry of continuing to race toward more advanced systems without sufficient safety precautions.
The proposal is likely to face significant opposition from the technology industry and others who argue that decisions about the development and use of AI should not be left solely to government regulators.
Zuckerberg frames AI superintelligence as the great democratizer
Meta CEO Mark Zuckerberg, for example, has argued that the central question surrounding superintelligence is who will have access to it rather than whether it should be developed at all.
Zuckerberg has also rejected the idea that a small group of experts should determine what is best for society. Instead, he argued that people should ultimately decide how AI tools are used, pointing to democracy and markets as mechanisms for allowing individuals to determine what constitutes the “best life.”
